Clinical De-identification - Portuguese

Input Format

To use the model, you need to provide input in one of the following supported formats:

JSON Format

Provide input as JSON. We support two variations within this format:

Array of Text Documents:

Use an array containing multiple text documents. Each element represents a separate text document.

{ "text": [ "Text document 1", "Text document 2", ... ] }

Single Text Document:

Provide a single text document as a string.

{ "text": "Single text document" }

JSON Lines (JSONL) Format

Provide input in JSON Lines format, where each line is a JSON object representing a text document.

{"text": "Text document 1"} {"text": "Text document 2"}

Important Parameter

masking_policy: str

Users can select a masking policy to determine how sensitive entities are handled:

Example:

"Dados do paciente. Nome: Mauro. Apelido: Gonçalves. NISS: 26 63514095. Endereço: Calle Miguel Benitez 90."

masked: Default policy that masks entities with their type.

-> 'Dados do . Nome: . Apelido: . NISS: . Endereço: .'

obfuscated: Replaces sensitive entities with random values of the same type.

-> 'Dados do Homen. Nome: Josefina Batista. Apelido: Tiago Santos. NISS: 16 63514095. Endereço: Avenida da Liberdade, 17.'

masked_fixed_length_chars: Masks entities with a fixed length of asterisks (*).

-> 'Dados do ****. Nome: ****. Apelido: ****. NISS: ****. Endereço: ****.'

masked_with_chars: Masks entities with asterisks (*).

-> 'Dados do []. Nome: []. Apelido: [*****]. NISS: []. Endereço: [************].'

You can specify these parameters in the input as follows:

{ "text": [ "Text document 1", "Text document 2", ... ], "masking_policy": "masked" }